1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
|
"""
=================================================
Partial Dependence Plots with categorical values
=================================================
Sigurd Carlsen Feb 2019
Holger Nahrstaedt 2020
.. currentmodule:: skopt
Plot objective now supports optional use of partial dependence as well as
different methods of defining parameter values for dependency plots.
"""
print(__doc__)
import numpy as np
from skopt.plots import plot_objective
np.random.seed(123)
import numpy as np
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import cross_val_score
from sklearn.tree import DecisionTreeClassifier
from skopt import gp_minimize
from skopt.plots import plot_objective
from skopt.space import Categorical, Integer
#############################################################################
# objective function
# ==================
# Here we define a function that we evaluate.
def objective(params):
clf = DecisionTreeClassifier(
**{dim.name: val for dim, val in zip(SPACE, params) if dim.name != 'dummy'}
)
return -np.mean(cross_val_score(clf, *load_breast_cancer(return_X_y=True)))
#############################################################################
# Bayesian optimization
# =====================
SPACE = [
Integer(1, 20, name='max_depth'),
Integer(2, 100, name='min_samples_split'),
Integer(5, 30, name='min_samples_leaf'),
Integer(1, 30, name='max_features'),
Categorical(list('abc'), name='dummy'),
Categorical(['gini', 'entropy'], name='criterion'),
Categorical(list('def'), name='dummy'),
]
result = gp_minimize(objective, SPACE, n_calls=20)
#############################################################################
# Partial dependence plot
# =======================
#
# Here we see an example of using partial dependence. Even when setting
# n_points all the way down to 10 from the default of 40, this method is
# still very slow. This is because partial dependence calculates 250 extra
# predictions for each point on the plots.
_ = plot_objective(result, n_points=10)
#############################################################################
# Plot without partial dependence
# ===============================
# Here we plot without partial dependence. We see that it is a lot faster.
# Also the values for the other parameters are set to the default "result"
# which is the parameter set of the best observed value so far. In the case
# of funny_func this is close to 0 for all parameters.
_ = plot_objective(result, sample_source='result', n_points=10)
#############################################################################
# Modify the shown minimum
# ========================
# Here we try with setting the other parameters to something other than
# "result". When dealing with categorical dimensions we can't use
# 'expected_minimum'. Therefore we try with "expected_minimum_random"
# which is a naive way of finding the minimum of the surrogate by only
# using random sampling. `n_minimum_search` sets the number of random samples,
# which is used to find the minimum
_ = plot_objective(
result,
n_points=10,
sample_source='expected_minimum_random',
minimum='expected_minimum_random',
n_minimum_search=10000,
)
#############################################################################
# Set a minimum location
# ======================
# Lastly we can also define these parameters ourselfs by
# parsing a list as the pars argument:
_ = plot_objective(
result,
n_points=10,
sample_source=[15, 4, 7, 15, 'b', 'entropy', 'e'],
minimum=[15, 4, 7, 15, 'b', 'entropy', 'e'],
)
|
